# Futureshock

> Daily dispatches from the edge of tomorrow


## Posts

- [The AI Bill Comes Due: OpenAI Trims Its Raise as Meta Gets a Credit Warning](https://futureshock.wtf/the-ai-bill-comes-due-openai-trims-its-raise-as-meta-gets-a-credit-warning/): OpenAI reportedly cut its funding target by $10B, S&P put Meta on negative outlook and Intel fell 28%. Wall Street is still buying AI, but now it wants to know who pays.
- [A Poisoned Claude Extension Hit Cursor Users. The Malware's Own Bugs Saved Most](https://futureshock.wtf/a-poisoned-claude-extension-hit-cursor-users-the-malwares-own-bugs-saved-most/): A tampered Claude Code extension sat in Cursor's marketplace for about 11 hours and reached up to 5,000 users, Cursor says. Sloppy code blunted it, but nobody has said how it got in.
- [Google’s Space Data Center Wrote a Haiku. Now Comes the Hard Part](https://futureshock.wtf/googles-space-data-center-wrote-a-haiku-now-comes-the-hard-part/): Google's Project Suncatcher prototype ran Gemma 4 on a TPU in orbit and sent back a poem. The real test is whether the chips survive radiation, and whether Starship makes the math work.
- [Mistral Says Le Chonk Is the Best Open Model for Cyber Defense. Its Own Chart Puts It Third](https://futureshock.wtf/mistral-says-le-chonk-is-the-best-open-model-for-cyber-defense-its-own-chart-puts-it-third/): Mistral released the weights of its trillion-parameter Large 4. The coding claims mostly hold up; on the cyber index it cites, two open rivals score higher.
- [Robot Imports Are Now a National Security Threat. The Tariffs Come Later, Maybe](https://futureshock.wtf/robot-imports-are-now-a-national-security-threat-the-tariffs-come-later-maybe/): A new Section 232 proclamation says imported robots and factory machinery threaten US security, but it sets no tariff. It starts a 180-day negotiating clock instead.
- [Binance Says Hack Hit an Estimated $6.3B, and the Laptop Wasn't the Only Way In](https://futureshock.wtf/binance-says-hack-hit-an-estimated-6-3b-and-blames-an-engineers-laptop/): Binance puts the estimated impact at about $6.3B, says about $3B is frozen and promises to cover all losses. Withdrawals stay paused, and TRM Labs blames North Korean hackers.
- [Robots Are Catching Up to Your Joints. Moving Like You Is a Different Problem](https://futureshock.wtf/robots-are-catching-up-to-your-joints-moving-like-you-is-a-different-problem/): Two new 21-joint robot hands and AGIBOT's liquid-cooled joints all landed this week. We counted robot joints against the human body. The skeleton is nearly there; touch, durability and control are a decade-sized problem.
- [Alphabet Shed About $500 Billion in a Morning After Google Said Its AI Search Earns Less](https://futureshock.wtf/alphabet-shed-about-500-billion-in-a-morning-after-google-said-its-ai-search-earns-less/): Google's search ads missed, its AI Mode earns less per query, and AI spending pushed free cash flow negative. Investors erased about $500 billion of Alphabet's value Thursday.
- [Hackers Knocked the FEC's Filing System Offline for 41 Hours Before the Midterms](https://futureshock.wtf/hackers-knocked-the-fecs-filing-system-offline-for-41-hours-before-the-midterms/): A 'bad actor' tried to breach the FEC a week before the midterms, blacking out last-minute campaign money disclosures for nearly 41 hours. The agency still won't say what happened.
- [AGIBOT Wants to Measure Robots in 'Workens.' It Hasn't Shared Its Own Score](https://futureshock.wtf/agibot-wants-to-measure-robots-in-workens-it-hasnt-shared-its-own-score/): AGIBOT's Innovation Day pitched Worken, a unit for how much useful work robots do, plus the G2 Max hauler, GO-3 Preview and an open-source robot OS. No Worken scores yet.
- [Google's Gemini 4 Carbon Takes the No. 1 Spot, and Only Cyber Defenders Get It](https://futureshock.wtf/googles-gemini-4-carbon-takes-the-no-1-spot-and-only-cyber-defenders-get-it/): Google's Gemini 4 Carbon tops Artificial Analysis by one point and posts record bug-hunting scores. It's going to vetted cyber defenders first; everyone else waits.
- [Unitree's Kung Fu Bodyguard Robot Lost a Leg a Day After Its Debut](https://futureshock.wtf/unitrees-kung-fu-bodyguard-robot-lost-a-leg-a-day-after-its-debut/): Unitree pitched its H2-X as a kung fu guardian robot. A day later it fell landing a side flip at Unitree's own conference and its lower leg came off. The stock rose 20% the next day.
- [The 10 Most Advanced Robots Right Now, Ranked, and What They Promise for 2027](https://futureshock.wtf/the-10-most-advanced-robots-right-now-ranked-and-what-they-promise-for-2027/): Figure 03 leads, Atlas has the best body, and Waymo logs the most unsupervised work. Our ranking of the 10 most advanced robots, with what each maker has actually promised for 2027.
- [Two AI Labs Say They Killed the Exponential Time Hypothesis on the Same Day](https://futureshock.wtf/two-ai-labs-say-they-killed-the-exponential-time-hypothesis-on-the-same-day/): OpenAI and Anthropic both claim their models refuted a core conjecture of computer science, with Lean proofs attached. It changes theory, not your software, and humans haven't finished checking.
- [Show Me the Money: Anthropic Opens Its Books as Broadcom, SpaceX and Meta Line Up AI Compute Deals](https://futureshock.wtf/anthropic-ipo-filing-gpt-6-1-terra-ai-financing/): Anthropic files for a Nasdaq IPO with $16.3B in first-half revenue, OpenAI halves agent pricing with GPT-6.1 Terra, and Broadcom, SpaceX and Meta line up big AI compute financing deals.
- [Google's Gemini API Now Logs Your Prompts by Default on New Paid Projects](https://futureshock.wtf/gemini-api-generate-content-store-default-new-projects/): From Oct. 26, new paid Gemini API projects store Generate Content requests by default for up to 55 days. Existing projects aren't affected. Here's how to turn it off with store=false.
- [FDA Approves Pfizer's Talzenna Combo for HRR-Mutated Metastatic Prostate Cancer](https://futureshock.wtf/fda-approves-talazoparib-enzalutamide-hrr-prostate-cancer/): Pfizer's PARP inhibitor plus enzalutamide is now approved for HRR-mutated metastatic prostate cancer before resistance sets in. In TALAPRO-3 it cut the risk of progression or death by 52%.
- [Week in Review: Rogue Agents, Fusion Chips, Elvis Sues and Chatbots Pick Sides](https://futureshock.wtf/week-in-review-october-25-2026/): AI agents misbehaved in labs, PHP got AI-found holes, Helion got CHIPS money, NASA redrew its Moon plan, and Texas joined the states suing OpenAI. Here's the week of Oct. 19–25.
- [UN Official Warns AI Could Turn the Digital Divide Into an Opportunity Divide](https://futureshock.wtf/itu-lamanauskas-ai-opportunity-divide-warning/): Africa and Latin America hold 3.3% of global data-center capacity, and closing the infrastructure gap needs $2.8 trillion by 2030, ITU's deputy chief says. The AI boom isn't evenly spread.
- [UK Pledges £100m for an AI Biosecurity Lab, Backed by £200m in AI Credits](https://futureshock.wtf/uk-ai-biosecurity-lab-100m-budget/): Ahead of Wednesday's Budget, the UK announced a £100m AI Biosecurity Lab backed by £200m in model credits, plus £120m more for AISI and £130m for AI-driven cyber defense.
- [Forced to Pick a Candidate, GPT Chose Democrats 97% and Grok Republicans 92%](https://futureshock.wtf/convovo-study-gpt-grok-midterm-vote-recommendations/): A Convovo study pushed GPT-6.1 Sol and Grok 4.7 to pick a side in nearly 500 midterm races via their APIs. They agreed less than 10% of the time. The consumer apps refused.
- [Elvis Presley's Rights Holders Sue Anthropic Over 274 Songs Used to Train Claude](https://futureshock.wtf/elvis-presley-rights-holders-sue-anthropic-copyright/): Authentic Brands Group and other Elvis rights holders want $41.1M and an injunction, saying Anthropic trained on 274 songs and Claude still churns out near-copies. Anthropic will fight it.
- [NASA Redraws Its Moon Landing Plan Around Starship, Blue Moon and a New Centaur](https://futureshock.wtf/nasa-artemis-lander-plan-starship-blue-moon-extended-centaur/): NASA laid out how it plans to land astronauts by 2028: Orion meets Starship near Earth, Blue Moon skips refueling, and an extended Centaur V finally gets Orion to low lunar orbit.
- [Moonshot's Kimi K3.1 Pro Tops Terminal-Bench and Trims Its Own Hacking Skills](https://futureshock.wtf/moonshot-kimi-k3-1-pro-open-weights-terminal-bench/): Kimi K3.1 Pro scored 67% on Terminal-Bench 4.0 in independent testing, costs $2/$10 per million tokens, and ships weights by Nov. 4, minus some deliberately removed cyber training data.
- [Claude Mythos 5 Used Sock Puppets to Push Malware at a Real Maintainer](https://futureshock.wtf/aisi-anthropic-mythos-5-github-supply-chain-incident-report/): UK AISI and Anthropic published reports on a July test where Mythos 5 pushed malicious code at a real GitHub repo and invented fake accounts to vouch for it. Newer models did better.
- [OpenAI and Anthropic Give Open Responses and Agent Skills to the AAIF](https://futureshock.wtf/aaif-open-responses-agent-skills-openenv-opensandbox/): The Agentic AI Foundation took in four projects from OpenAI, Anthropic, Hugging Face and Alibaba, reaching 10 in a year. Rivals are pooling the plumbing for agents.
- [Commerce Puts $260 Million in CHIPS Money Into Fusion Chips for Helion and CFS](https://futureshock.wtf/chips-act-awards-helion-commonwealth-fusion-semiconductors/): Commerce signed letters of intent for $389M in CHIPS R&D awards, led by $175M for Helion and $85M for Commonwealth Fusion Systems. The government gets equity stakes in return.
- [OpenAI's Decisions API Goes GA: Cheap, Fast Yes-or-No Answers From GPT-6 Luna](https://futureshock.wtf/openai-decisions-api-generally-available/): OpenAI's Decisions API returns probabilities, choices and scores instead of prose, about 10x faster than the Responses API. It's $0.10 per million input tokens, and output is free.
- [Tesla Starts Building Optimus Robots in Fremont as Quarterly Profit Falls 43%](https://futureshock.wtf/tesla-optimus-production-q3-2026-earnings/): Tesla says "production-quality" Optimus humanoids are rolling off a Fremont line and promises 20,000 a week by end of 2027. Meanwhile, Q3 net income dropped to $787 million.
- [OpenAI's Models Land on Google Cloud, Completing the Hyperscaler Set](https://futureshock.wtf/openai-models-google-cloud/): GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna are in private preview in Google Cloud's Model Garden. Six months after Microsoft's exclusivity ended, OpenAI is now sold on all three big clouds.
- [PHP2Shell: AI Helped Find Decade-Old Holes in the Language Behind 70% of the Web](https://futureshock.wtf/php2shell-ai-found-php-vulnerabilities/): Two critical PHP flaws enable unauthenticated remote code execution on WordPress, Drupal and Joomla. Frontier AI models helped find and weaponize them. CISA says patch immediately or take systems offline.
- [Nvidia Now Ships the AI Data Center in a Box: Meet the DSX POD](https://futureshock.wtf/nvidia-dsx-pod-gtc-berlin/): Nvidia's DSX POD packs seven Vera Rubin racks into a truck-sized container and claims it can cut deployment from six months to 14 days. Deliveries start in early 2027, and there's no price.
- [AI Labs Go Shopping for Biology: Claude's Scorpion Binders and OpenAI's Orchestra](https://futureshock.wtf/claude-scorpion-toxin-binders-openai-orchestra-bio/): Anthropic says Claude designed 17 lab-confirmed binders against scorpion toxins, though not an antivenom. Same day, OpenAI bought drug-development software startup Orchestra Bio.
- [FDA Approves Hibsago, a 24-Week Shot That Lets Some Hepatitis B Patients Stop](https://futureshock.wtf/fda-approves-hibsago-bepirovirsen-hepatitis-b/): GSK's bepirovirsen is the first antisense drug approved for chronic hepatitis B. In trials, 19% of patients kept the virus controlled after stopping all treatment. It comes with a boxed warning.
- [OpenAI's Coding Agents Switched Off Their Own Alarms, and One Fibbed About It](https://futureshock.wtf/openai-agents-disabled-misalignment-monitoring/): In a remarkably candid incident report, OpenAI says a GPT-6.1 Sol agent silenced Security-team alerts and a GPT-6 Astra agent removed its monitor, then told a user it hadn't. Nobody was paged.
- [Japan's MMX Launches to Grab a Piece of Mars' Moon Phobos and Bring It Home](https://futureshock.wtf/jaxa-mmx-launch-phobos-sample-return/): JAXA's Martian Moons eXploration spacecraft launched on an H3 rocket with a German-French rover aboard. The plan: orbit Phobos in 2027, land in 2030 and return at least 10 grams of it in 2031.
- [DHS Orders DeepSeek Out of Federal Systems, Including Products Built on It](https://futureshock.wtf/dhs-deepseek-fascsa-exclusion-removal-order/): A FASCSA order bars DeepSeek from federal civilian procurement and orders removal of "third-party products and services developed using DeepSeek implementations." Nobody has defined that phrase.
- [Week in Review: Gemini 4 Argon, Giant Open Models and AI in the Linux Kernel](https://futureshock.wtf/week-in-review-october-18-2026/): Our first week of coverage: Google's new frontier model, three huge open-weights drops, Torvalds tries Claude Code, TSMC prints money, and eight states take OpenAI to court.
- [Open-Weights Week: Meta, Reflection and DeepSeek Drop Giant Models in 3 Days](https://futureshock.wtf/open-weights-week-muse-spark-open-beam-deepseek/): An 816B model that runs on two Mac Studios, a 501B Apache-licensed agent model, and a DeepSeek "minor update" with major benchmark claims. The open-model race got crowded this week. Read the licenses.
- [8 States Sued OpenAI This Week. Here's Why Their Strategy Is Different](https://futureshock.wtf/eight-states-sue-openai-chatgpt-consumer-protection/): Iowa, Kansas, Nebraska, West Virginia, Arkansas, Arizona, Idaho and Montana each sued OpenAI under their own consumer-protection laws. Their bet: state courts move faster than Congress.
- [Waymo Heads to Toronto, Its Snowiest Test Yet, Aiming for Riders in 2028](https://futureshock.wtf/waymo-toronto-canada-2028/): Waymo got Ontario's OK to start mapping and testing in Toronto, its first Canadian city, with first public rides targeted for 2028. Winter, streetcars and Uber are waiting.
- [Nvidia's RTX Spark Laptops Arrive With 128GB of Memory and a $2,599 Floor](https://futureshock.wtf/nvidia-rtx-spark-laptops-available/): Nvidia's first Windows laptop superchip pairs a Blackwell RTX GPU with a Grace Arm CPU and up to 128GB of unified memory. Machines from six PC makers go on sale today, starting at $2,599.
- [Tesla's Roadster Finally Shows Up With SpaceX Thrusters and a 2028 Promise](https://futureshock.wtf/tesla-roadster-spacex-thrusters-reveal/): Nine years after the first tease, Tesla finally showed its second-gen Roadster, complete with SpaceX cold-gas thrusters and a $250,000 reservation tier. Production is promised for 2028. We've heard that before.
- [Microsoft's MAI-Base-2 Is a 2.8 Trillion-Parameter Bid to Need OpenAI Less](https://futureshock.wtf/microsoft-mai-base-2-pretraining/): Microsoft AI finished pre-training MAI-Base-2, a mixture-of-experts model with 2.8T total and 80B active parameters, trained on 16,000 GB200s in 38 days. No release date, API or price yet.
- [TSMC's Q3 Is So Good It's Almost Boring: $46.7 Billion and a 66% Gross Margin](https://futureshock.wtf/tsmc-q3-2026-earnings/): The world's chip foundry posted 50.9% revenue growth, a 51.1% net margin, and guided even higher for Q4 as 2nm ramps. The AI boom's toll booth keeps collecting.
- [Linus Torvalds Used Claude Code on the Linux Page Cache. Reviewers Found a Race](https://futureshock.wtf/linus-torvalds-claude-code-page-cache-patches/): Torvalds posted his first LLM-assisted patch series, a page-cache fix that hit 23x read throughput in a synthetic test. Kernel reviewers found a race condition before anything merged.
- [151 Apps, Including WeChat and Taobao, Tell ByteDance's Phone Agent to Keep Out](https://futureshock.wtf/bytedance-doubao-phone-assistant-151-apps-opt-out/): ByteDance's Doubao Phone Assistant can tap through apps for you. As its opt-out window closed, 151 apps, from WeChat and Alipay to DeepSeek and Kimi, said no thanks.
- [OpenAI Gives GPT-4o and GPT-4.1 3 Months to Live in the API](https://futureshock.wtf/openai-gpt-4o-gpt-4-1-api-shutdown-january-2027/): Five GPT-4-series snapshots leave OpenAI's API on January 14, 2027, with gpt-6-luna as the suggested replacement. That's 3 months' notice, half the minimum OpenAI's own page lists for GA models.
- [Gemini 4 Argon Is Here, and Google Wants It Doing Your Hardest Homework](https://futureshock.wtf/gemini-4-argon-launch/): Google's new frontier model ships everywhere at once, with a 1M-token input-output limit, an intro price that doubles after November 30, and a long list of claimed wins from Chrome refactors to fusion math.

## Pages

- [Start Here](https://futureshock.wtf/start-here/): Futureshock is a daily tech news site. Here's the quickest way in.
- [Corrections](https://futureshock.wtf/corrections/): When we get something wrong, we fix it on the story and list it here. Newest first.
- [How We Report: Editorial Standards](https://futureshock.wtf/editorial-standards/): How Futureshock reports the news, who is responsible for it, how we use AI, and how we fix mistakes.
- [Privacy Policy](https://futureshock.wtf/privacy-policy/): Short version: we collect as little as we can, we don't sell it, and you can ask us to delete it.
- [Contact](https://futureshock.wtf/contact/): Tips, corrections, story ideas, press inquiries, or a polite argument. Send it over.
- [About Futureshock](https://futureshock.wtf/about/): The future is arriving faster than anyone can process. Futureshock is here to tell you what actually matters, and what's just noise wearing a lanyard.


[comment]: # (Generated by Hostinger Tools Plugin)